Application of human-in-the-loop hybrid augmented intelligence approach in security inspection system

A security inspection system exemplifies human-machine collaboration, and enhancing its safety and reliability through advanced technology remains a key research priority. While deep learning has incrementally improved the autonomous capabilities of security inspection equipment for automatic contraband detection, a gap persists between current technological capabilities and practical implementation. Recognizing that humans excel at learning, reasoning, and collaborating, while artificial intelligence offers normative, repeatable, and logical processing, we propose a human-in-the-loop hybrid augmented intelligence approach. This approach addresses the practical needs of security inspection systems by introducing a hybrid decision-making method that leverages two distinct strategies: "Reject-priority" and "Clear-priority." These strategies play complementary roles in bolstering the decision-making process's overall performance. Comparative experiments on a dataset from a specific security inspection site confirmed the hybrid method's effectiveness, drawing several conclusions. This "Hybrid decision-making" method not only enhances risk perception, thereby widening the safety margin of the security inspection system, but also reduces the need for human labor, leading to increased efficiency and reduced labor costs. Additionally, it is less time-consuming, further improving the system's overall efficiency. By integrating human and machine intelligence, this method significantly boosts decision-making effectiveness. Tailored to their unique characteristics, the method based on "Reject-priority" strategy is particularly well-suited for security inspection scenarios that demand stringent safety protocols, while the "Clear-priority" method is ideal for scenarios with high-volume traffic flow, where efficiency is paramount. As the volume of collected data grows, this approach will enable seamless adaptation of the method to evolving application needs.